MAREANO - NiN - hi Dens below 1500m depth

Packages

library(magrittr)

package 㤼㸱magrittr㤼㸲 was built under R version 4.0.3

library(dplyr)

package 㤼㸱dplyr㤼㸲 was built under R version 4.0.3
Attaching package: 㤼㸱dplyr㤼㸲

The following objects are masked from 㤼㸱package:stats㤼㸲:

    filter, lag

The following objects are masked from 㤼㸱package:base㤼㸲:

    intersect, setdiff, setequal, union

library(ggcorrplot)

package 㤼㸱ggcorrplot㤼㸲 was built under R version 4.0.5Loading required package: ggplot2
package 㤼㸱ggplot2㤼㸲 was built under R version 4.0.5Need help? Try Stackoverflow: https://stackoverflow.com/tags/ggplot2

library(ggpubr)

package 㤼㸱ggpubr㤼㸲 was built under R version 4.0.3Registered S3 method overwritten by 'data.table':
  method           from
  print.data.table     

library(patchwork)

package 㤼㸱patchwork㤼㸲 was built under R version 4.0.5

# library("here")
# library("bookdown")
# library("downloadthis")
library(vegan)

package 㤼㸱vegan㤼㸲 was built under R version 4.0.5Loading required package: permute
Loading required package: lattice
This is vegan 2.5-7

library(plyr)

---------------------------------------------------------------------------------------------------------------------------------------
You have loaded plyr after dplyr - this is likely to cause problems.
If you need functions from both plyr and dplyr, please load plyr first, then dplyr:
library(plyr); library(dplyr)
---------------------------------------------------------------------------------------------------------------------------------------

Attaching package: 㤼㸱plyr㤼㸲

The following object is masked from 㤼㸱package:ggpubr㤼㸲:

    mutate

The following objects are masked from 㤼㸱package:dplyr㤼㸲:

    arrange, count, desc, failwith, id, mutate, rename, summarise, summarize

library(e1071)

package 㤼㸱e1071㤼㸲 was built under R version 4.0.3

library(tidyverse)

package 㤼㸱tidyverse㤼㸲 was built under R version 4.0.3Registered S3 methods overwritten by 'dbplyr':
  method         from
  print.tbl_lazy     
  print.tbl_sql      
-- Attaching packages ------------------------------------------------------------------------------------------------ tidyverse 1.3.0 --
√ tibble  3.0.4     √ purrr   0.3.4
√ tidyr   1.1.2     √ stringr 1.4.0
√ readr   1.4.0     √ forcats 0.5.0
package 㤼㸱tibble㤼㸲 was built under R version 4.0.3package 㤼㸱tidyr㤼㸲 was built under R version 4.0.3package 㤼㸱readr㤼㸲 was built under R version 4.0.3package 㤼㸱purrr㤼㸲 was built under R version 4.0.5-- Conflicts --------------------------------------------------------------------------------------------------- tidyverse_conflicts() --
x plyr::arrange()    masks dplyr::arrange()
x purrr::compact()   masks plyr::compact()
x plyr::count()      masks dplyr::count()
x tidyr::extract()   masks magrittr::extract()
x plyr::failwith()   masks dplyr::failwith()
x dplyr::filter()    masks stats::filter()
x plyr::id()         masks dplyr::id()
x dplyr::lag()       masks stats::lag()
x plyr::mutate()     masks ggpubr::mutate(), dplyr::mutate()
x plyr::rename()     masks dplyr::rename()
x purrr::set_names() masks magrittr::set_names()
x plyr::summarise()  masks dplyr::summarise()
x plyr::summarize()  masks dplyr::summarize()

library(viridis)

Loading required package: viridisLite

library(GGally)

package 㤼㸱GGally㤼㸲 was built under R version 4.0.5Registered S3 method overwritten by 'GGally':
  method from   
  +.gg   ggplot2

library(ggrepel)

package 㤼㸱ggrepel㤼㸲 was built under R version 4.0.3

library(ggordiplots)

Loading required package: glue

Attaching package: 㤼㸱glue㤼㸲

The following object is masked from 㤼㸱package:dplyr㤼㸲:

    collapse

library(readr)
library(RColorBrewer)

package 㤼㸱RColorBrewer㤼㸲 was built under R version 4.0.5

library(oce) 

package 㤼㸱oce㤼㸲 was built under R version 4.0.5Loading required package: gsw
package 㤼㸱gsw㤼㸲 was built under R version 4.0.5

library(plotly)

package 㤼㸱plotly㤼㸲 was built under R version 4.0.5Registered S3 methods overwritten by 'htmltools':
  method               from         
  print.html           tools:rstudio
  print.shiny.tag      tools:rstudio
  print.shiny.tag.list tools:rstudio
Registered S3 method overwritten by 'htmlwidgets':
  method           from         
  print.htmlwidget tools:rstudio

Attaching package: 㤼㸱plotly㤼㸲

The following objects are masked from 㤼㸱package:plyr㤼㸲:

    arrange, mutate, rename, summarise

The following object is masked from 㤼㸱package:ggplot2㤼㸲:

    last_plot

The following object is masked from 㤼㸱package:stats㤼㸲:

    filter

The following object is masked from 㤼㸱package:graphics㤼㸲:

    layout

library(purrr)
library(furrr)

Loading required package: future

source("gg_ordisurf_viridis.R")

LOAD OBJECTS IF HAVE RUN ALREADY

geodist_pa<-readRDS(file.path(dataPath,"inputs/SPLITS/DEEP/hiDens_b1500_geodist_pa.rds"))

cannot open compressed file 'C:/Users/a21448/Havforskningsinstituttet/NiN + MAREANO - General/inputs/SPLITS/DEEP/hiDens_b1500_geodist_pa.rds', probable reason 'No such file or directory'Error in gzfile(file, "rb") : cannot open the connection

Data

env_sort <- read.csv(file.path(dataPath,"inputs/SPLITS/DEEP/DeepBelow1500_env_sort_2022-09-30.csv")) %>% as.data.frame
otu_sort <- read.csv(file.path(dataPath,"inputs/SPLITS/DEEP/DeepBelow1500_otu_sort_2022-09-30.csv")) %>% as.data.frame


table(env_sort$SampID2 == otu_sort$SampID)

< table of extent 0 >

#must be same length and equal to unique sample number length - the ordered lists are assumed to be directly relatable on a row by row basis

dim(env_sort)

[1] 190 138

dim(otu_sort)

[1] 190 391

remove any variables with not enough coverage

env_sort <- env_sort %>% select(-c("BO22_lightbotmean_bdmean",
                               "BO22_lightbotltmax_bdmean",
                               "BO22_lightbotltmin_bdmean",
                               "BO22_lightbotrange_bdmean"))

Error in app$vspace(new_style$`margin-top` %||% 0) : 
  attempt to apply non-function

Plot map to check location of samples

env_sort_locations<- ggplot(data = env_sort,
                      aes(x = X.y,
                          y = Y)) +
  theme_classic() +
  geom_point(aes(colour = bathy),
             size = 2) +
  scale_colour_gradient2(low = "red",
                         mid = "yellow",
                         high = "green") +
  ggtitle("Location of samples in sorted env file")

env_sort_locations

Data cleaning

## Removing NAs
otuCompl <- otu_sort[complete.cases(env_sort[, -c(1:which(colnames(env_sort)=="bathy")-1)]), ]
envCompl <- env_sort[complete.cases(env_sort[, -c(1:which(colnames(env_sort)=="bathy")-1)]), ]

## Removing observations with less than 4 OTUs
sel <- rowSums(otuCompl[, -c(1:2)]) >= 4
otuSel <- otuCompl[sel, ]
envSel <- envCompl[sel, ]


## Removing phosphate
#envSel <- envSel %>% select(-phosphate_mean.tif)


dim(otuSel); dim(envSel)

[1] 161 391
[1] 161 134

Show what samples are left after complete cases and >4 OTUs filters

env_sel_locations<- ggplot(data = envSel,
                      aes(x = X.y,
                          y = Y)) +
  theme_classic() +
  geom_point(aes(colour = bathy),
             size = 2) +
  scale_colour_gradient2(low = "red",
                         mid = "yellow",
                         high = "green") +
  ggtitle("Location of samples in sorted env file")

env_sel_locations

Show what got removed in complete cases filter

envCompl_ccrem<-env_sort%>%filter(!SampID%in%envCompl$SampID)

envCompl_ccrem_locations<- ggplot(data = envCompl_ccrem,
                      aes(x = X.y,
                          y = Y)) +
  theme_classic() +
  geom_point(aes(colour = bathy),
             size = 2) +
  scale_colour_gradient2(low = "red",
                         mid = "yellow",
                         high = "green") +
  ggtitle("Location of removed complete case samples resulting in envSel file")

envCompl_ccrem_locations

Show what got removed in <4 OTUs filter

inv.sel <- rowSums(otuCompl[, -c(1:2)]) < 4
env.invSel <- envCompl[inv.sel, ]

env.invSel_locations<- ggplot(data = env.invSel,
                      aes(x = X.y,
                          y = Y)) +
  theme_classic() +
  geom_point(aes(colour = bathy),
             size = 2) +
  scale_colour_gradient2(low = "red",
                         mid = "yellow",
                         high = "green") +
  ggtitle("Location of removed samples with <4 OTUs resulting in envSel file")

env.invSel_locations

[Optional] Data thinning

# otu_red <- otu1[1:500, ]
# otu <- otu_red
# 
# env_red <- env[1:500, ]
# env <- env_red

Whole dataset for first run

otu<-otuSel[,-c(1:which(colnames(otuSel)=="Acesta_excavata")-1)]
env<-envSel[, -c(2:(which(colnames(env_sort)=="bathy")-1))]

table(is.na(otu))

FALSE 
62629 

table(is.na(env))

FALSE 
18837 

#str(otuSel)
#str(envSel)

Splitting in subsets

# ## Class 1
# otu1 <- subset(otuSel, envSel$SplitRev == 1)
# env1 <- envSel %>% filter(SplitRev == 1)
# 
# ## Class 2
# otu2 <- subset(otuSel, envSel$SplitRev == 2)
# env2 <- envSel %>% filter(SplitRev == 2)
# 
# ## Class 4
# otu3 <- subset(otuSel, envSel$SplitRev == 3)
# env3 <- envSel %>% filter(SplitRev == 3)
# 
# ## Class 4
# otu4 <- subset(otuSel, envSel$SplitRev == 4)
# env4 <- envSel %>% filter(SplitRev == 4)
# 
# ## Class 6
# otu6 <- subset(otuSel, envSel$SplitRev == 6)
# env6 <- envSel %>% filter(SplitRev == 6)
# 
# ## Class 7
# otu7 <- subset(otuSel, envSel$SplitRev == 7)
# env7 <- envSel %>% filter(SplitRev == 7)
# 
# ## Class 8
# otu8 <- subset(otuSel, envSel$SplitRev == 8)
# env8 <- envSel %>% filter(SplitRev == 8)

Selecting subset

# ## Selecting 
# otu <- otu2
# env <- env2

format data correctly

#env$coords.x1<-as.numeric(env$coords.x1)
#env$coords.x2<-as.numeric(env$coords.x2)

Abundance weighting

Make function You might have to drop variables that have been imported as character

otu_pa <- decostand(x = otu[, -c(1)],
                    method = "pa")

# y = ax^w         # power transformation formula
dt <- otu[, -c(1:2)]          # species data to transform
x_mn <- min(dt[dt > 0])
x_mx <- max(dt)
rng <- 6           # abundance range
w <- log(rng) / (log(x_mx) - log(x_mn))
a <- x_mn^(-w)
# otu_6 <- a * dt[, -c(1:3)]^w
otu_6 <- a * dt^w
range(otu_6)

[1] 0 6

Ordination methods

DCA PA

Sys.time()

[1] "2022-10-03 12:14:10 CEST"

dca_pa <- decorana(veg = otu_pa)

some species were removed because they were missing in the data

print(dca_pa, head=T)

Call:
decorana(veg = otu_pa) 

Detrended correspondence analysis with 26 segments.
Rescaling of axes with 4 iterations.

                  DCA1   DCA2   DCA3   DCA4
Eigenvalues     0.4412 0.3659 0.2158 0.3282
Decorana values 0.4811 0.4035 0.3339 0.3026
Axis lengths    4.3022 3.5983 3.3305 4.2530

GNMDS PA

Distances - don’t run if loading saved object

## Bray-Curtis
dist_pa <- vegdist(x = otu_pa, method = "bray")

## Geodist
ep <- 0.8     # epsilon
geodist_pa <- isomapdist(dist = dist_pa, epsilon = ep)

Save the result - don’t run if loading saved object

saveRDS(geodist_pa,
        file = (file.path(dataPath,"inputs/SPLITS/DEEP/hiDens_b1500_b1500_geodist_pa.rds")))

Ordination

monoMDS - don’t run if loading saved object

took 6.4mins with jonatan’s paralell solution (usually ~10mins)

Using old version without furrr

 Sys.time()

[1] "2022-09-30 15:06:20 CEST"

 d <- 2
 mds_pa <- list()

 for (i in 1:100) {
   mds_pa[[i]]<-monoMDS(geodist_pa,
                     matrix(c(runif(dim(otu_pa)[1]*d)),
                            nrow = dim(otu_pa)[1]),
                     k = d,
                     model = "global",
                     maxit = 2000,
                     smin = 1e-7,
                     sfgrmin = 1e-7)
 }
 Sys.time()

[1] "2022-09-30 15:06:33 CEST"

# # monoMDS
#  d <- 2
# mds_r6 <- list()
# 
#  Sys.time()
#  for (i in 1:200) {
#    mds_r6[[i]]<-monoMDS(geodist_r6,
#                      matrix(c(runif(dim(otu_6)[1]*d)),
#                             nrow = dim(otu_6)[1]),
#                      k = d,
#                      model = "global",
#                      maxit = 2000,
#                      smin = 1e-7,
#                      sfgrmin = 1e-7)# }
#  Sys.time()

# #Jonatan's upgrade for speed - parallel processing of mds reptitions
# library(furrr) # Package to run non sequential functions in parallel
# #library(purrr)
# plan(multisession)
# 
# d <- 2
# 
# i <-200 # number of reps
# 
# List_geodist_pa <- lapply(seq_len(i), function(X) geodist_pa) # makes a list with the input data repeated as many times as reps wanted
# start_t<-Sys.time()
# Xmds_pa<-furrr::future_map(List_geodist_pa, 
#                   function(x) monoMDS(x,
#                                       matrix(c(runif(dim(otu_6)[1]*d)),
#                                              nrow = dim(otu_6)[1]),
#                                       k = d,
#                                       model = "global",
#                                       maxit = 2000,
#                                       smin = 1e-7,
#                                       sfgrmin = 1e-7),
#                   .progress = TRUE)
# Sys.time() - start_t

Save the result - don’t run if loading saved object

can take a few mins

saveRDS(mds_pa,
        file = file.path(dataPath,"inputs/SPLITS/DEEP/hiDens_b1500_mds_pa.rds"))

Best nmds solution - PA

Make function?

# Loading geodist object
# geodist_nfi <- readRDS(file = "../SplitRev2_geodist_pa_full.rds")

# Loading mds results
# mds <- readRDS(file = "../SplitRev2_mds_pa_full.rds")

## Extracting the stress of each nmds iteration
mds_stress_pa<-unlist(lapply(mds_pa, function(v){v[[22]]})) 

ordered_pa <-order(mds_stress_pa)

## Best, second best, and worst solution
mds_stress_pa[ordered_pa[1]]

[1] 0.2601212

mds_stress_pa[ordered_pa[2]]

[1] 0.2603306

mds_stress_pa[ordered_pa[10]]

[1] 0.2608846

## Scaling of axes to half change units and varimax rotation by postMDS
mds_best_pa<-postMDS(mds_pa[[ordered_pa[1]]],
                  geodist_pa, 
                  pc = TRUE, 
                  halfchange = TRUE, 
                  threshold = ep)     # Is this threshold related to the epsilon above?
mds_best_pa

Call:
monoMDS(dist = geodist_pa, y = matrix(c(runif(dim(otu_pa)[1] *      d)), nrow = dim(otu_pa)[1]), k = d, model = "global", maxit = 2000,      smin = 1e-07, sfgrmin = 1e-07) 

Non-metric Multidimensional Scaling

161 points, dissimilarity ‘bray shortest isomap’, call ‘isomapdist(dist = dist_pa, epsilon = ep)’

Dimensions: 2 
Stress:     0.2601212 
Stress type 1, weak ties
Scores scaled to unit root mean square, rotated to principal components
Stopped after 104 iterations: Stress nearly unchanged (ratio > sratmax)

mds_secbest_pa <- postMDS(mds_pa[[ordered_pa[2]]],
                          geodist_pa, 
                          pc = TRUE, 
                          halfchange = TRUE, 
                          threshold = ep)
mds_secbest_pa

Call:
monoMDS(dist = geodist_pa, y = matrix(c(runif(dim(otu_pa)[1] *      d)), nrow = dim(otu_pa)[1]), k = d, model = "global", maxit = 2000,      smin = 1e-07, sfgrmin = 1e-07) 

Non-metric Multidimensional Scaling

161 points, dissimilarity ‘bray shortest isomap’, call ‘isomapdist(dist = dist_pa, epsilon = ep)’

Dimensions: 2 
Stress:     0.2603306 
Stress type 1, weak ties
Scores scaled to unit root mean square, rotated to principal components
Stopped after 248 iterations: Stress nearly unchanged (ratio > sratmax)

## Procrustes comparisons
procr_pa <- procrustes(mds_best_pa,
                    mds_secbest_pa,
                    permutations=999)
protest(mds_best_pa,
        mds_secbest_pa,
        permutations=999)

Call:
protest(X = mds_best_pa, Y = mds_secbest_pa, permutations = 999) 

Procrustes Sum of Squares (m12 squared):        0.1237 
Correlation in a symmetric Procrustes rotation: 0.9361 
Significance:  0.001 

Permutation: free
Number of permutations: 999

plot(procr_pa)

png(file=file.path(dataPath,"outputs/hiDens_b1500_procrustes_pa.png"), width=1000, height=700)

plot(procr_pa)
dev.off()

png 
  2 

Correlation of axis: DCA vs NMDS - PA

# Extracting ordination axis
ax <- 2
axis_pa <- cbind(mds_best_pa$points,
                 scores(dca_pa,
                        display = "sites",
                        origin = TRUE)[, 1:ax])

ggcorr(axis_pa, 
       method=c("everything","kendall"), 
       label = TRUE,
       label_size = 3, 
       label_color = "black",  
       nbreaks = 8,
       label_round = 3,
       low = "red",
       mid = "white",
       high = "green")

NA
NA

Save the figure

ggsave(filename = file.path(dataPath,"outputs/hiDens_b1500_correlationPCAvsNMDS_PA.png"),
       device = "png",
       dpi=300 )

Saving 7 x 7 in image

# Switching direction of NMDS1
# mds_best$points[, 1] <- -mds_best$points[, 1]

DCA R6

dca_r6 <- decorana(veg = otu_6)

some species were removed because they were missing in the data

print(dca_r6, head=T)

Call:
decorana(veg = otu_6) 

Detrended correspondence analysis with 26 segments.
Rescaling of axes with 4 iterations.

                  DCA1   DCA2   DCA3   DCA4
Eigenvalues     0.4624 0.3966 0.3092 0.3106
Decorana values 0.5075 0.4233 0.3731 0.2867
Axis lengths    4.3845 4.1435 3.1853 3.4650

GNMDS R6

Distances - don’t run if loading saved object

## Bray-Curtis
dist_r6 <- vegdist(x = otu_6, method = "bray")

## Geodist
ep <- 0.80     # epsilon
geodist_r6 <- isomapdist(dist = dist_r6, epsilon = ep)

Save the result - don’t run if loading saved object

saveRDS(geodist_r6,
        file = (file.path(dataPath,"inputs/SPLITS/DEEP/hiDens_b1500_geodist_r6.rds")))

Ordination

200 reps - don’t run if loading saved object

(Jonatan’s parallel solution took 5.4mins, before it took 10mins)

Using older version without furrr.

## monoMDS
d <- 2
mds_r6 <- list()

Sys.time()

[1] "2022-09-30 15:07:13 CEST"

for (i in 1:200) {
  mds_r6[[i]]<-monoMDS(geodist_r6,
                    matrix(c(runif(dim(otu_6)[1]*d)),
                           nrow = dim(otu_6)[1]),
                    k = d,
                    model = "global",
                    maxit = 2000,
                    smin = 1e-7,
                    sfgrmin = 1e-7)
}
Sys.time()

[1] "2022-09-30 15:07:28 CEST"

# #Jonatan's upgrade for speed - parallel processing of mds reptitions
# #library(furrr) # Package to run non sequential functions in parallel
# #library(purrr)
# plan(multisession)
# 
# d <- 2
# 
# i <-200 # number of reps
# 
# List_geodist_r6 <- lapply(seq_len(i), function(X) geodist_r6) # makes a list with the input data repeated as many times as reps wanted
# start_t<-Sys.time()
# Xmds_r6<-furrr::future_map(List_geodist_r6, 
#                   function(x) monoMDS(x,
#                                       matrix(c(runif(dim(otu_6)[1]*d)),
#                                              nrow = dim(otu_6)[1]),
#                                       k = d,
#                                       model = "global",
#                                       maxit = 2000,
#                                       smin = 1e-7,
#                                       sfgrmin = 1e-7),
#                   .progress = TRUE)
# Sys.time() - start_t

Save the result - don’t run if loading saved object

saveRDS(mds_r6,
        file = file.path(dataPath,"inputs/SPLITS/DEEP/hiDens_b1500_mds_r6.rds")) 

Best nmds solution r6 200 rep

# Loading geodist object
# geodist_nfi <- readRDS(file = "../SplitRev2_geodist_nfi.rds")

# Loading mds results
# mds <- readRDS(file = "../SplitRev2_mds.rds")

## Extracting the stress of each nmds iteration
mds_stress_r6<-unlist(lapply(mds_r6, function(v){v[[22]]})) 

ordered_r6 <-order(mds_stress_r6)

## Best, second best, and worst solution
mds_stress_r6[ordered_r6[1]]

[1] 0.2524937

mds_stress_r6[ordered_r6[2]]

[1] 0.2525011

mds_stress_r6[ordered_r6[100]]

[1] 0.2581464

## Scaling of axes to half change units and varimax rotation by postMDS
mds_best_r6<-postMDS(mds_r6[[ordered_r6[1]]],
                     geodist_r6, 
                     pc = TRUE, 
                     halfchange = TRUE, 
                     threshold = ep)     # Is this threshold related to the epsilon above?
mds_best_r6

Call:
monoMDS(dist = geodist_r6, y = matrix(c(runif(dim(otu_6)[1] *      d)), nrow = dim(otu_6)[1]), k = d, model = "global", maxit = 2000,      smin = 1e-07, sfgrmin = 1e-07) 

Non-metric Multidimensional Scaling

161 points, dissimilarity ‘bray shortest isomap’, call ‘isomapdist(dist = dist_r6, epsilon = ep)’

Dimensions: 2 
Stress:     0.2524937 
Stress type 1, weak ties
Scores scaled to unit root mean square, rotated to principal components
Stopped after 70 iterations: Stress nearly unchanged (ratio > sratmax)

mds_secbest_r6<-postMDS(mds_r6[[ordered_r6[2]]],
                        geodist_r6, 
                        pc = TRUE, 
                        halfchange = TRUE, 
                        threshold = ep)
mds_secbest_r6

Call:
monoMDS(dist = geodist_r6, y = matrix(c(runif(dim(otu_6)[1] *      d)), nrow = dim(otu_6)[1]), k = d, model = "global", maxit = 2000,      smin = 1e-07, sfgrmin = 1e-07) 

Non-metric Multidimensional Scaling

161 points, dissimilarity ‘bray shortest isomap’, call ‘isomapdist(dist = dist_r6, epsilon = ep)’

Dimensions: 2 
Stress:     0.2525011 
Stress type 1, weak ties
Scores scaled to unit root mean square, rotated to principal components
Stopped after 118 iterations: Stress nearly unchanged (ratio > sratmax)

## Procrustes comparisons
procr_r6 <- procrustes(mds_best_r6,
                       mds_secbest_r6,
                       permutations=999)
protest(mds_best_r6,
        mds_secbest_r6,
        permutations=999)

Call:
protest(X = mds_best_r6, Y = mds_secbest_r6, permutations = 999) 

Procrustes Sum of Squares (m12 squared):        0.000444 
Correlation in a symmetric Procrustes rotation: 0.9998 
Significance:  0.001 

Permutation: free
Number of permutations: 999

plot(procr_r6)

png(file.path(dataPath,"outputs/hiDens_b1500_procrustes_r6.png"), width=1000, height=700,) #added 1000

plot(procr_r6)
dev.off()

png 
  2 

#### 1000 reps - don’t run if loading saved object Currently commented out as it gained nothing but took extra time. Can be removed in due course.

Correlation of axis: DCA vs NMDS

retain the 200 rep version

# Extracting ordination axis
ax <- 2
axis_r6 <- cbind(mds_best_r6$points,
                 scores(dca_r6,
                        display = "sites",
                        origin = TRUE)[, 1:ax])

ggcorr(axis_r6, 
       method=c("everything","kendall"), 
       label = TRUE,
       label_size = 3, 
       label_color = "black",  
       nbreaks = 8,
       label_round = 3,
       low = "red",
       mid = "white",
       high = "green")

Save the figure

ggsave(filename = file.path(dataPath,"outputs/hiDens_b1500_correlationPCAvsNMDS_r6.png"),
       device = "png",
       dpi=300 )

Saving 7 x 7 in image

# Switching direction of NMDS1
# mds_best$points[, 1] <- -mds_best$points[, 1]

Plotting DCA & GNMDS

## Adding scores to data frame
otu_6$gnmds1 <- mds_best_r6$points[, 1]
otu_6$gnmds2 <- mds_best_r6$points[, 2]
otu_6$dca1 <- scores(dca_r6, display = "sites", origin = TRUE)[, 1]
otu_6$dca2 <- scores(dca_r6, display = "sites", origin = TRUE)[, 2]

p_gnmds_r6 <- ggplot(data = otu_6,
                     aes(x = gnmds1,
                         y = gnmds2)) +
  theme_classic() +
  coord_fixed() +
  ggtitle("GNMDS",
          subtitle = "First run") +
  geom_point(colour = "red") +
  geom_vline(xintercept = 0,
             linetype = 2,
             colour = "lightgray") +
  geom_hline(yintercept = 0,
             linetype = 2,
             colour = "lightgray")

p_dca_r6 <- ggplot(data = otu_6,
                   aes(x = dca1,
                       y = dca2)) +
  theme_classic() +
  coord_fixed() +
  ggtitle("DCA",
          subtitle = "First run") +
  geom_point(colour = "red") +
  geom_vline(xintercept = 0,
             linetype = 2,
             colour = "lightgray") +
  geom_hline(yintercept = 0,
             linetype = 2,
             colour = "lightgray")

p_gnmds_r6 + p_dca_r6

NA
NA

Save the figure

ggsave(file.path(dataPath,"outputs/hiDens_b1500_gnmds_dca_r6.png"),
       device = "png", 
       dpi=300)

Saving 7 x 7 in image

Species-environment relationships

Selecting ordination

ord <- mds_best_r6

## Axis scores if selected ord is GNMDS
axis <- ord$points %>% as.data.frame

## Axis scores if selected ord is DCA
# axis <- scores(ord,
#                display = "sites",
#                origin = TRUE)[, 1:ax])

Create additional variables

decided to make MLD-bathy vars

env<-env %>% 
  mutate ("MLDmean_bathy"=MLDmean_Robinson-(bathy*-1),
          "MLDmin_bathy"=MLDmin_Robinson-(bathy*-1),
          "MLDmax_bathy"=MLDmax_Robinson-(bathy*-1))

env$MLDmean_bathy<-cut(env$MLDmean_bathy, 
      breaks=c(-2560, -20,20,130),#checked range of values first (min -2554, max 123)
      labels=c('belowMLD','onPycno','inMixLayer'))
env$MLDmin_bathy<-cut(env$MLDmin_bathy, 
      breaks=c(-2560, -20,20,130),#checked range of values first (min -2554, max 123)
      labels=c('belowMLD','onPycno','inMixLayer'))
env$MLDmax_bathy<-cut(env$MLDmax_bathy, 
      breaks=c(-2560, -20,20,130),#checked range of values first (min -2554, max 123)
      labels=c('belowMLD','onPycno','inMixLayer'))

env$swDensRob_avs<-swRho(salinity=env$Smean_Robinson,
                         temperature=env$Tmean_Robinson,
                         pressure=(env$bathy*-1),
                         eos="unesco")

Correlation ordination axes and environmental variables

Removing non-env vars

env_cont<-env%>% select(-c(landscape,sedclass,gmorph, MLDmean_bathy, MLDmax_bathy, MLDmin_bathy, #categorical
                           optional, #not a var
                           MLDmax_Robinson, MLDmean_Robinson, MLDmin_Robinson,  #replaced by new vars
                           MLDsd_Robinson #not meaningful
                           ))
env_cont<-env_cont%>% mutate_if(is.integer,as.numeric)
env_corr <- env_cont #%>% select(-c(SampID))

# env_corr$coords.x1<-as.numeric(env_corr$coords.x1)
# env_corr$coords.x2<-as.numeric(env_corr$coords.x2)

env_corr[(!is.numeric(env_corr)),]

Correlations

# Vector to hold correlations
cor_ax1 <- NULL
cor_ax2 <- NULL
pv_ax1 <- NULL
pv_ax2 <- NULL

# NMDS1
for( i in seq(length(env_corr))) {
  ct.i <- cor.test(axis$MDS1,
                   env_corr[, i],
                   method = "kendall")
  cor_ax1[i] <- ct.i$estimate
  pv_ax1[i] <- ct.i$p.value
}

the standard deviation is zerothe standard deviation is zerothe standard deviation is zerothe standard deviation is zerothe standard deviation is zerothe standard deviation is zero

# NMDS2
for( i in seq(length(env_corr))) {
  ct.i <- cor.test(axis$MDS2,
                   env_corr[, i],
                   method = "kendall")
  cor_ax2[i] <- ct.i$estimate
  pv_ax2[i] <- ct.i$p.value
}

the standard deviation is zerothe standard deviation is zerothe standard deviation is zerothe standard deviation is zerothe standard deviation is zerothe standard deviation is zero

cor_tab <- data.frame(env = names(env_corr),
                      ord_ax1 = cor_ax1,
                      pval_ax1 = pv_ax1,
                      ord_ax2 = cor_ax2,
                      pval_ax2 = pv_ax2)

cor_tab

write.csv(x = cor_tab,
          file = file.path(dataPath,"inputs/SPLITS/DEEP/hiDens_b1500_cor-table_r6_200rep_MLD-bathy.csv"),
          row.names = FALSE)

Dot chart to check for gaps in correlation

cor_a1_sort<-cor_tab%>%

There were 15 warnings (use warnings() to see them)

  mutate(abs_ord_ax1=abs(ord_ax1),
         abs_ord_ax2=abs(ord_ax2)) %>%
  arrange(desc(abs_ord_ax1))

cor_a2_sort<-cor_tab%>%
  mutate(abs_ord_ax1=abs(ord_ax1),
         abs_ord_ax2=abs(ord_ax2)) %>%
  arrange(desc(abs_ord_ax2))

dotchart(cor_a1_sort$abs_ord_ax1, main="Absolute (+/-) correlations between envVars and gnmds axis 1")

cor_cut<-0.50 #decide

cor_sel<-subset(cor_a1_sort,abs_ord_ax1>cor_cut)
cor_sel

as.data.frame(cor_sel$env)

Sel env var (top corr)

env_os <- env[, cor_sel$env]
env_os

str(env_os)

'data.frame':   161 obs. of  11 variables:
 $ swDensRob_avs               : num  1037 1037 1036 1036 1035 ...
 $ bathy                       : num  -1961 -2009 -1598 -1596 -1594 ...
 $ Tmean_Robinson              : num  -0.246 -0.251 0.45 0.528 0.607 ...
 $ Smax_Robinson               : num  35 34.9 35 35 35 ...
 $ Tmax_Robinson               : num  -0.234 -0.234 0.498 0.579 0.66 ...
 $ Smean_Robinson              : num  34.9 34.9 35 35 35 ...
 $ Tmin_Robinson               : num  -0.271 -0.299 0.382 0.458 0.534 ...
 $ Smin_Robinson               : num  34.9 34.9 35 35 35 ...
 $ BO22_carbonphytoltmin_bdmean: num  0.0199 0.0199 0.0197 0.0197 0.0197 ...
 $ BO22_ironmean_bdmean        : num  0.000684 0.000687 0.000674 0.000673 0.000673 ...
 $ BO22_ironrange_bdmean       : num  0.000684 0.000687 0.000674 0.000673 0.000673 ...

Ordisurfs top corr

ordsrfs <- list(length = ncol(env_os))

for (i in seq(ncol(env_os))) {
  os.i <- gg_ordisurf(ord = ord,
                      env.var = env_os[, i],
                      pt.size = 1,
                      # binwidth = 0.05,
                      var.label = names(env_os)[i],
                      gen.text.size = 10,
                      title.text.size = 15,
                      leg.text.size = 10
                      )
  
  ordsrfs[[i]] <- os.i$plot
}

ordsrfs_plt <- ggarrange(plotlist = ordsrfs,
                         nrow = 4,
                         ncol = 3)

ordsrfs_plt

Save some outputs

ggexport(ordsrfs_plt,
          filename = file.path(dataPath,"outputs/hiDens_b1500_ordisurfs_top_corr.png"),
          width = 1000,
          height = 1000)

Sel env var (manual)

env_os_m <- env[,c("Tmean_Robinson", #top corr
                  "salt_max", #top corr
                  "Smax_Robinson", #comparison to top corr
                  "swDensRob_avs", #top corr
                  "BO22_icecoverltmax_ss",#top corr ax2
                  "BO22_icecovermean_ss",#top corr
                  "BO22_dissoxmean_bdmean",#top corr
                  #"BO22_carbonphytoltmin_bdmean",#top corr - no clear gradient in ordisurf
                  "BO22_ppltmin_ss", #top corr
                  "X.y", #comparison to Y
                  "Y", #top corr
                  "spd_std", #top corr ax2 (blended model)
                  "CSpdsd_Robinson", #comparison to top corr ax2 (blended model)
                  "mud", #highest sed var ax1 + corr
                  "gravel",#highest sed var ax1 - corr
                  "BO22_silicateltmax_bdmean", #just under top corr ax1
                  "bathy", #intuitive for comparisons
                  "slope9"#intuitive for comparisons
                )]
env_os_m

str(env_os_m)

'data.frame':   161 obs. of  17 variables:
 $ Tmean_Robinson           : num  -0.246 -0.251 0.45 0.528 0.607 ...
 $ salt_max                 : num  34.9 34.9 35 35 35 ...
 $ Smax_Robinson            : num  35 34.9 35 35 35 ...
 $ swDensRob_avs            : num  1037 1037 1036 1036 1035 ...
 $ BO22_icecoverltmax_ss    : num  9.78e-05 9.71e-05 0.00 0.00 0.00 ...
 $ BO22_icecovermean_ss     : num  1.22e-05 1.25e-05 0.00 0.00 0.00 ...
 $ BO22_dissoxmean_bdmean   : num  304 304 293 293 293 ...
 $ BO22_ppltmin_ss          : num  0.00 0.00 1.49e-05 1.49e-05 1.49e-05 ...
 $ X.y                      : num  463061 470861 554861 555061 555261 ...
 $ Y                        : num  8311934 8295534 7744134 7744134 7744134 ...
 $ spd_std                  : num  0.0511 0.0508 0.0726 0.0736 0.0748 ...
 $ CSpdsd_Robinson          : num  0.014566 0.01478 0.000513 0.000492 0.00047 ...
 $ mud                      : num  96 96 69.5 69.5 69.5 0 0 80 0 0 ...
 $ gravel                   : num  0 0 0.5 0.5 0.5 0 0 0 0 0 ...
 $ BO22_silicateltmax_bdmean: num  13.2 13.4 10.8 10.8 10.8 ...
 $ bathy                    : num  -1961 -2009 -1598 -1596 -1594 ...
 $ slope9                   : num  0.814 0.672 3.516 3.255 3.39 ...

Ordisurfs manually selected

ordsrfs_m <- list(length = ncol(env_os_m))

for (i in seq(ncol(env_os_m))) {
  os.i_m <- gg_ordisurf(ord = ord,
                      env.var = env_os_m[, i],
                      pt.size = 1,
                      # binwidth = 0.05,
                      var.label = names(env_os_m)[i],
                      gen.text.size = 10,
                      title.text.size = 15,
                      leg.text.size = 10)
  
  ordsrfs_m[[i]] <- os.i_m$plot
}

ordsrfs_plt_m <- ggarrange(plotlist = ordsrfs_m,
                         nrow = 6,
                         ncol = 3)

ordsrfs_plt_m

Save some outputs

ggexport(ordsrfs_plt_m,
          filename = file.path(dataPath,"outputs/hiDens_b1500_ordisurfs_man_sel_domean.png"),
          width = 2000,
          height = 2000)

Envfit

## Select if any var should be excluded from envfit (makes less busy to read)
env_os_m_envfit<-env_os_m [,c("Tmean_Robinson", #top corr
                  "salt_max", #top corr
                  "Smax_Robinson", #comparison to top corr
                  "swDensRob_avs", #top corr
                  "BO22_icecoverltmax_ss",#top corr ax2
                  #"BO22_icecovermean_ss",#top corr
                  "BO22_dissoxmean_bdmean",#top corr
                  #"BO22_dissoxltmin_bdmean",#top corr
                  #"BO22_carbonphytoltmin_bdmean",#top corr - no clear gradient in ordisurf
                  "BO22_ppltmin_ss", #top corr
                  "X.y", #comparison to Y
                  "Y", #top corr
                  "spd_std", #top corr ax2 (blended model)
                 # "CSpdsd_Robinson", #comparison to top corr ax2 (blended model)
                  "mud", #highest sed var ax1 + corr
                  "gravel",#highest sed var ax1 - corr
                  "BO22_silicateltmax_bdmean", #just under top corr ax1
                  "bathy" #intuitive for comparisons
                
  
)]

colnames(env_os_m_envfit)<-c("T", #top corr
                  "sMx", #top corr
                  "SmaxR", #comparison to top corr
                  "swDensR", #top corr
                  "icecovmax",#top corr ax2
                  #"icecovav",#top corr
                  "dissoxav",#top corr
                  #"dissoxmin",#top corr
                  #"BO22_carbonphytoltmin_bdmean",#top corr - no clear gradient in ordisurf
                  "ppltmin", #top corr
                  "X", #comparison to Y
                  "Y", #top corr
                  "spd_std", #top corr ax2 (blended model)
                 # "CSsd", #comparison to top corr ax2 (blended model)
                  "mud", #highest sed var ax1 + corr
                  "gravel",#highest sed var ax1 - corr
                  "SiLtmax", #just under top corr ax1
                  "bathy" #intuitive for comparisons
                 )

## Envfot plot
gg_envfit(ord = ord,
          env = env_os_m_envfit,
          pt.size = 1)

## Envfit analysis

ef <- envfit(ord = ord,
             env = env_os_m_envfit,
             # na.rm = TRUE
             )
efDF <- as.data.frame(scores(ef,
                             display = "vectors"))

Save the plot

ggsave(filename = file.path(dataPath,"outputs/hiDens_b1500_EnvFit_man_sel_cln_domean.png"),
       device = "png",
       dpi=300 )

Saving 7 x 7 in image

Categorical envVar visualised on the mdsplots

use dataset with categorical var included

env_vis<-env
env_vis$gnmds1 <- otu_6$gnmds1
env_vis$gnmds2 <- otu_6$gnmds2

env_vis$dca1 <- otu_6$dca1
env_vis$dca2 <- otu_6$dca2

env_vis$SampID <- envSel$SampID

gnmds w mld mean - bathy

dca_int <- ggplot(data = env_vis,
                     aes(x = dca1,
                         y = dca2)) +
  theme_classic() +
  coord_fixed() +
  ggtitle("Interactive DCA sample ID plot",
          subtitle = "First run") +
  geom_point(aes(colour = factor(SampID))) +
  geom_vline(xintercept = 0,
             linetype = 2,
             colour = "lightgray") +
  geom_hline(yintercept = 0,
             linetype = 2,
             colour = "lightgray")

ggplotly(dca_int)

gnmds w mld mean - bathy

p_mld <- ggplot(data = env_vis,
                     aes(x = gnmds1,
                         y = gnmds2)) +
  theme_classic() +
  coord_fixed() +
  ggtitle("GNMDS coloured by proximity to mixed layer depth",
          subtitle = "First run") +
  geom_point(aes(colour = MLDmean_bathy)) +
  geom_vline(xintercept = 0,
             linetype = 2,
             colour = "lightgray") +
  geom_hline(yintercept = 0,
             linetype = 2,
             colour = "lightgray")

p_mld

gnmds w sedclass

deal with sedclass codes

env_vis<- env_vis %>% 
  mutate(
    sedclassName = case_when(
            sedclass == "1" ~ "SedCoverR",
            sedclass == "5" ~ "Rock",
            sedclass == "20" ~ "Mud",
            sedclass == "21" ~ "MwBlock",
            sedclass == "40" ~ "sMud",
            sedclass == "80" ~ "mSand",
            sedclass == "100" ~ "Sand",
            sedclass == "110" ~ "gMud",
            sedclass == "115" ~ "gsMud",
            sedclass == "120" ~ "gmSand",
            sedclass == "130" ~ "gSand",
            sedclass == "150" ~ "MSG",
            sedclass == "160" ~ "sGravel",
            sedclass == "170" ~ "Gravel",
            sedclass == "175" ~ "GravBlock",
            sedclass == "185" ~ "SGBmix",
            sedclass == "205" ~ "S/MwB",
            sedclass == "206" ~ "S/MwG/B",
            sedclass == "215" ~ "SGBalt",
            sedclass == "300" ~ "HardSed",
            sedclass == "500" ~ "Biogenic"
                        
    )
  )

colour palette to cope with up to 25 categorical colours

c25 <- c(
  "dodgerblue2", "#E31A1C", # red
  "green4",
  "#6A3D9A", # purple
  "#FF7F00", # orange
  "black", "gold1",
  "skyblue2", "#FB9A99", # lt pink
  "palegreen2",
  "#CAB2D6", # lt purple
  "#FDBF6F", # lt orange
  "gray70", "khaki2",
  "maroon", "orchid1", "hiDens_b1500pink1", "blue1", "steelblue4",
  "darkturquoise", "green1", "yellow4", "yellow3",
  "darkorange4", "brown"
)

p_sed <- ggplot(data = env_vis,
                     aes(x = gnmds1,
                         y = gnmds2)) +
  theme_classic() +
  coord_fixed() +
  ggtitle("GNMDS coloured by sediment class",
          subtitle = "First run") +
  geom_point(aes(colour = factor(sedclassName))) +
   scale_colour_manual(values=c25)+
  geom_vline(xintercept = 0,
             linetype = 2,
             colour = "lightgray") +
  geom_hline(yintercept = 0,
             linetype = 2,
             colour = "lightgray")+
  guides(colour=guide_legend(ncol=2))

p_sed

gnmds w lanscape

deal with sedclass codes

env_vis<- env_vis %>% 
  mutate(
    landscapeName = case_when(
            landscape == "1" ~ "Strandflat",
            landscape == "21" ~ "ContSlope",
            landscape == "22" ~ "Canyon",
            landscape == "31" ~ "Valley",
            landscape == "32" ~ "Fjord",
            landscape == "41" ~ "DeepSeaPlane",
            landscape == "42" ~ "SlopePlain",
            landscape == "43" ~ "ShelfPlain",
            landscape == "431" ~ "shallowValley"
    )
  )

p_land <- ggplot(data = env_vis,
                     aes(x = gnmds1,
                         y = gnmds2)) +
  theme_classic() +
  coord_fixed() +
  ggtitle("GNMDS coloured by landscape class",
          subtitle = "First run") +
  geom_point(aes(colour = factor(landscapeName))) +
  geom_vline(xintercept = 0,
             linetype = 2,
             colour = "lightgray") +
  geom_hline(yintercept = 0,
             linetype = 2,
             colour = "lightgray")+
  guides(colour=guide_legend(ncol=2))

p_land

Gnmds w gmorph

p_gmo <- ggplot(data = env_vis,
                     aes(x = gnmds1,
                         y = gnmds2)) +
  theme_classic() +
  coord_fixed() +
  ggtitle("GNMDS coloured by landscape class",
          subtitle = "First run") +
  geom_point(aes(colour = factor(gmorph))) +
  geom_vline(xintercept = 0,
             linetype = 2,
             colour = "lightgray") +
  geom_hline(yintercept = 0,
             linetype = 2,
             colour = "lightgray")+
  guides(colour=guide_legend(ncol=2))

p_gmo

cat_var_plots<-p_mld+p_sed+p_land+p_gmo

Save the plot

ggexport(cat_var_plots,
          filename = file.path(dataPath,"outputs/hiDens_b1500_gnmds_catvar.png"),
          width = 1000,
          height = 800)

Sample identification in the mdsplot

p_gmo <- ggplot(data = env_vis,
                     aes(x = gnmds1,
                         y = gnmds2)) +
  theme_classic() +
  coord_fixed() +
  ggtitle("GNMDS coloured by sample",
          subtitle = "First run") +
  geom_point(aes(colour = factor(SampID))) +
  geom_vline(xintercept = 0,
             linetype = 2,
             colour = "lightgray") +
  geom_hline(yintercept = 0,
             linetype = 2,
             colour = "lightgray")+
  guides(colour="none", size="none")

ggplotly(p_gmo)